Fundraise

KAIST professors' AI inference startup AnyBridge lands seed round from The Ventures

What's the deal? AnyBridgeDealroom has a profile for this one. Try Dealroom →, a South Korean AI infrastructure software startup, has closed a seed round led by The VenturesDealroom has a profile for this one. Try Dealroom →, the firm announced on October 22, 2026. The amount was not disclosed. The company was also selected for TIPSDealroom has a profile for this one. Try Dealroom →, a government program that backs the deal with up to US$670.7K in research funding over two years.

What does it build? AnyBridge develops a "multi-accelerator runtime" that lets companies run large language models across different AI chips — GPUs, NPUs, and PIM — rather than GPUs alone. A common interface automatically assigns inference tasks to the best-suited chip, targeting a 40–70% cut in inference costs versus GPU-only setups.

Who's behind it? The company was co-founded by KAIST computer science professors Park Jong-se and Kwon Young-jin, who together hold 26 AI-system patents and have published more than 110 papers. Kunle Olukotun, Stanford professor and SambaNova co-founder, serves as an adviser. Its technology builds on LLMServingSim, an open-source serving simulator that won a best-paper award at IISWC 2024.

Why now? The generative AI market is shifting from model-performance races toward operating efficiency — cost per token and throughput. Alternative chips like NPUs and PIM have emerged as cheaper, more power-efficient options, but the software to deploy them in live LLM services remains early-stage.

What's the endgame? AnyBridge plans a formal product launch in 2027, offered in licensed, managed, and usage-based "token factory" forms. It aims to serve customers from AI startups to public, financial, and enterprise clients where data security matters.

"GPU-focused LLM environments can be expanded to a range of AI chips," said The Ventures chief executive officer Kim Chul-woo, adding the team should "significantly lower companies' AI adoption costs."

What's next? The funds will go toward product development and customer validation. AnyBridge aims to run pilots with two to three customers this year and a proof-of-concept with Kakao Cloud. It has already signed development agreements with domestic AI chipmakers FuriosaAI and HyperAccel.

The signal: Korea's AI infrastructure race is widening beyond large-scale GPU and data-center buildouts toward software and specialised hardware that lower real-world inference costs. Against moves like Rebellions' large funding rounds and government-led GPU supply, this looks like a small, early-stage bet on a credible technical team.

Read more: wowtale.net

Image credit: cbowns

More top stories